test: replace util.inherits with es6 extends

[bnoordhuis]

I wanted to see how hard it would be to eradicate util.inherits() from an existing code base and I figured, having done the work, I might as well PR it.

I did a quick and partial attempt at updating lib/ as well but it makes for huge diffs, plus it breaks code that calls constructors as functions (e.g. server = http.Server(cb).)

$ git stash show lib/_debug_agent.js | 237 ++++++++++++++++++------------------
lib/_debugger.js | 808 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++--------------------------------------------------------------
lib/_http_agent.js | 460 +++++++++++++++++++++++++++++++++++----------------------------------
lib/_http_client.js | 590 ++++++++++++++++++++++++++++++++++++++++++++---------------------------------------------
lib/_http_incoming.js | 275 +++++++++++++++++++++---------------------
lib/http.js | 85 ++++++-------
lib/https.js | 253 +++++++++++++++++++-------------------
test/parallel/test-http-keepalive-client.js | 4 +-
test/parallel/test-http-keepalive-maxsockets.js | 4 +-
test/parallel/test-http-keepalive-request.js | 4 +-
test/parallel/test-https-agent-servername.js | 2 +-
test/parallel/test-https-agent-sni.js | 2 +-
test/parallel/test-https-agent.js | 2 +-
13 files changed, 1347 insertions(+), 1379 deletions(-)

CI: https://ci.nodejs.org/job/node-test-pull-request/2477/

[eljefedelrodeodeljefe]

There is also one occurrence in test/common.js, which does not break.

[bnoordhuis]

I got that one in the first commit, didn't I? Or did I overlook something?

[eljefedelrodeodeljefe]

Ah sorry. Was just looking at your stash output. My bad.

[cjihrig]

LGTM. Any plans for a breaking change to lib?

[jasnell]

Not a fan of changing the existing test cases to this for two reasons:

1. It makes them significantly more difficult to backport to LTS if we need to
2. It could potentially introduce subtle differences in the test that might impact the results of the test.

I don't mind the change in the benchmark or test/common.js tho

[Fishrock123]

It makes them significantly more difficult to backport to LTS if we need to

Agreed, unless there are tangible positives to this I fail to see a good reason for now. (maybe in a year or something?)

[bnoordhuis]

Guess I'll close this, then.

It makes them significantly more difficult to backport to LTS if we need to

Not if this PR got backported, right? Or am I misunderstanding you?

It could potentially introduce subtle differences in the test that might impact the results of the test.

Shrug. I don't think that's likely. I wouldn't have opened the PR if I did.